indication that the change in fundamental frequency was causally related
to changes in formants due to mouth opening, and other time-synched
analyses of formant changes with mouth opening indicate that no causal
connection is likely (Hauser et al. 1993; see Fitch and Hauser 1995).
Thus, there is a significant body of data indicating independence of source
and filter in many vertebrate species. Nonetheless, independence is best
considered a working hypothesis at present, given our limited knowledge
of animal vocal production. Although source/filter theory is well-tested and
well-accepted in speech acoustics, independence of source and filter is only
a first-order approximation even in speech, and some interactions between
the two do occur (Bickley and Stevens 1986; Mergell and Herzel 1997).
Early in vitro work in birds provided some evidence of strong source/tract
coupling. For example, Rüppel (1933) found that the vibratory frequency
of an excised crane syrinx was dependent on the length of the vocal tract
attached suprasyringeally. Furthermore, a more recent paper looking at in
vivo production showed no heliox effect on intact birds but a profound
effect on a budgerigar with a denervated syrinx (Brittan-Powell et al. 1997).
These data suggest that source and filter may be passively coupled in this
species but that the bird normally overrides this coupling via active control
of the syrinx. This hypothesis would explain both the in vitro results of
Rüppel (1933) and the lack of evidence for source/tract coupling in the vast
majority of more recent studies. Another possible type of source/tract coupling, suggested by Hartley and Suthers (1988) for an echolocating bat
species, is that energy propagating back from the trachea could provide
positive feedback to support high-amplitude phonation, which is critical to
receiving a sufficiently strong echo from their echolocation cries. Despite
its plausibility, this hypothesis remains speculative at present. Finally, there
are many bird and mammal species for which the fundamental frequency
is close to the predicted formant frequencies based on vocal tract length,
suggesting the possibility for source/tract coupling, but whose production
has not been experimentally examined.
In conclusion, the accumulated data for terrestrial vertebrates, direct and
indirect, suggest that independence of source and filter should be assumed
as the working hypothesis of researchers in vertebrate bioacoustics as it is
in human speech. Specific data (e.g., derived from vocalizations in heliox)
would have to be adduced before rejecting this hypothesis and positing
source/tract coupling. This is worth stressing because many physicists and
bioacousticians, particularly in the older literature, adopt wind instruments,
and therefore coupling between source and filter, as their default model of
acoustic production. All current data suggest that the wind instrument
analogy is dangerously misleading as a model of vocal production. Moreover, independence of source and filter has an important practical consequence for researchers interested in studying call perception: using
well-developed techniques from speech science such as linear prediction
(LPC) or cepstral modeling, it is possible to pull a signal apart into source
84
W.T. Fitch and M.D. Hauser
to changes in formants due to mouth opening, and other time-synched
analyses of formant changes with mouth opening indicate that no causal
connection is likely (Hauser et al. 1993; see Fitch and Hauser 1995).
Thus, there is a significant body of data indicating independence of source
and filter in many vertebrate species. Nonetheless, independence is best
considered a working hypothesis at present, given our limited knowledge
of animal vocal production. Although source/filter theory is well-tested and
well-accepted in speech acoustics, independence of source and filter is only
a first-order approximation even in speech, and some interactions between
the two do occur (Bickley and Stevens 1986; Mergell and Herzel 1997).
Early in vitro work in birds provided some evidence of strong source/tract
coupling. For example, Rüppel (1933) found that the vibratory frequency
of an excised crane syrinx was dependent on the length of the vocal tract
attached suprasyringeally. Furthermore, a more recent paper looking at in
vivo production showed no heliox effect on intact birds but a profound
effect on a budgerigar with a denervated syrinx (Brittan-Powell et al. 1997).
These data suggest that source and filter may be passively coupled in this
species but that the bird normally overrides this coupling via active control
of the syrinx. This hypothesis would explain both the in vitro results of
Rüppel (1933) and the lack of evidence for source/tract coupling in the vast
majority of more recent studies. Another possible type of source/tract coupling, suggested by Hartley and Suthers (1988) for an echolocating bat
species, is that energy propagating back from the trachea could provide
positive feedback to support high-amplitude phonation, which is critical to
receiving a sufficiently strong echo from their echolocation cries. Despite
its plausibility, this hypothesis remains speculative at present. Finally, there
are many bird and mammal species for which the fundamental frequency
is close to the predicted formant frequencies based on vocal tract length,
suggesting the possibility for source/tract coupling, but whose production
has not been experimentally examined.
In conclusion, the accumulated data for terrestrial vertebrates, direct and
indirect, suggest that independence of source and filter should be assumed
as the working hypothesis of researchers in vertebrate bioacoustics as it is
in human speech. Specific data (e.g., derived from vocalizations in heliox)
would have to be adduced before rejecting this hypothesis and positing
source/tract coupling. This is worth stressing because many physicists and
bioacousticians, particularly in the older literature, adopt wind instruments,
and therefore coupling between source and filter, as their default model of
acoustic production. All current data suggest that the wind instrument
analogy is dangerously misleading as a model of vocal production. Moreover, independence of source and filter has an important practical consequence for researchers interested in studying call perception: using
well-developed techniques from speech science such as linear prediction
(LPC) or cepstral modeling, it is possible to pull a signal apart into source
84
W.T. Fitch and M.D. Hauser